Code smells and bogosity and fixes for ascent guidance optimizer

[lamont-granquist]

• The Phases are tightly coupled to the Optimizer. If anything, they should be more tightly coupled to the Problem.
• Ascent initial guess needs to be moved out of the Optimizer (need to have Phases decoupled from Optimizer first, though).
• The Optimizer shouldn't be setting u0 on the phases (can u0 come from the Solution instead? I think I'm using that for something janky to do with unguided stages, and removing the ODE integrator may have greatly reduced the need for the jank?)
• Need more box boundaries in the Optimizer
• Need inequality constraint to force positive coast times in the Optimizer
• Break the reliance on a fixed number of variables per phase in transcription (guided stages don't need all the control constraints)
• Look at fine-tuning how old Solutions generate new initial guesses to reduce Optimizer work and accelerate convergence.